Breast Masses Classification using a Sparse Representation

Breast mass detection and classification in mammograms is considered a very difficult task in medical image analysis. In this paper, we present a novel approach for classification of masses in digital mammograms according with their severity (benign or malign). Unlike other approaches, we do not segment masses but instead, we attempt to describe entire regions of interest (RoIs) based on a sparse representation. A set of patches selected by a radiologist in a RoI are characterized by their projection onto learned dictionaries, constructed previously from classified regions. Finally, the region class was identified using a decision rule algorithm. The strategy was assessed in a set of 80 masses with different shapes extracted from the DDSM database. The classification was compared with a ground truth already provided in the data base, showing an average accuracy rate of 70%.